When hinting glyphs, round y-offsets to integers

[tomcur]

Without this, while glyph outlines will be hinted to integers, the translation can bump the outlines to fractional locations again.

[waywardmonkeys]

This makes me think of rounding stuff in Parley recently.

[tomcur]

This makes me think of rounding stuff in Parley recently.

This one? linebender/parley#297

[waywardmonkeys]

Yes, sir.

[DJMcNab]

I don't know if we have any good docs for this, but https://docs.google.com/document/d/1gEqf7ehTzd89Djf_VpkL0B_Fb15e0w5fuv_UzyacAPU/edit?disco=AAABctut6QY is relevant (it's a link to a specific comment) here.

I don't think this is quite the right place to be doing this quantisation; we would need to be rounding the absolute y coordinate, not the "local" y coordinate.

Separately: ideally, there's be some way to warn if this changed frame-by-frame (that is, the baseline isn't rounded). But I don't think there's any way to detect that, because the relationship between y and the baseline is not so simple.

[tomcur]

I don't think this is quite the right place to be doing this quantisation; we would need to be rounding the absolute y coordinate, not the "local" y coordinate.

Hmm... I guess you're right, in that depending on GlyphRenderer, we don't quite know where the outline is going to be placed. I believe currently using vello_cpu, the y-coordinate here is the absolute coordinate.

[DJMcNab]

Hmm... I guess you're right, in that depending on GlyphRenderer, we don't quite know where the outline is going to be placed. I believe currently using vello_cpu, the y-coordinate here is the absolute coordinate.

Is it not impacted by the "adding" of the transform on (new) line 163?

[tomcur]

Is it not impacted by the "adding" of the transform on (new) line 163?

I was going to write it is not, but on second thought it is partially impacted. The order of operations is effectively:

transform * translate * skew * glyph_transform

(Where you have to read b * a as b after a.) In this code-path transform is guaranteed to have a scale of 1 because take_uniform_scale has taken the scale out of transform.

However, if transform itself translates with fractional coordinates, the outline can still be bumped to a fractional coordinate. We can account for that here.

[taj-p]

LGTM